GPU advancements in M3 and A17 Pro [video]
Apple’s new GPU architecture in the M3 chips and A17 Pro iPhone processor is drawing attention for offering unusually detailed public information about shader cores, dynamic register allocation, and ray tracing, aimed at helping developers better exploit Metal on Apple hardware. Commenters debate how technical the material really is versus marketing, compare Apple’s core and FLOPS metrics to Nvidia and AMD GPUs, and explore broader questions like portability to Linux, the role of tools like Apple’s Game Porting Toolkit, and how these changes may reduce issues such as shader compilation stutter. Some also highlight practical concerns, from narration style and accessibility of the content to lingering hardware quirks like external monitor flicker on earlier Apple Silicon Macs.
Perceived technical value of the video
- Some viewers initially felt it sounded like marketing, but many say it becomes quite technical after the first few minutes.
- It’s framed as an overview for a broad developer audience, not GPU designers, so detail is constrained but still considered unusually deep for a vendor talk.
- Several commenters say it clearly explains new M3/A17 GPU features, tradeoffs, and how to optimize Metal code for them.
GPU architecture details & comparisons
- The talk is praised for describing the old vs new shader core organization and Apple’s ray tracing acceleration, including register pressure and dynamic allocation behavior.
- Apple “GPU cores” are compared conceptually to Nvidia “SMs”; rough mappings to CUDA cores are discussed but seen as inherently fuzzy.
- Many argue core counts are misleading across vendors; FLOPs and actual bottlenecks (bandwidth, occupancy, register file, etc.) matter more.
Dynamic registers / caching
- A key theme is Apple’s move toward dynamic register allocation from a cache-like pool, described as a “magic workaround” to the traditional static register/occupancy tradeoff.
- Deeper questions (latency, exact implementation) are marked as unknown; likely microarchitectural and not documented.
Metal, DirectX, and game ecosystems
- One commenter suggests Apple should adopt DirectX; others reply it’s Windows‑tied, closed, and unlikely on macOS.
- A Proton-like translation layer and Apple’s “Game Porting Toolkit” are seen as more realistic paths.
- Debate over whether more games run on Metal vs modern DirectX; counts and definitions (mobile vs PC/console) are disputed and left unclear.
Apple vs competitors (SoCs & strategy)
- Some praise Apple’s CPU/GPU efficiency and criticize Google/Qualcomm designs as more complex and less efficient per core.
- Others emphasize Apple’s success coming from focus, tight integration, and ability to buy cutting‑edge nodes, not just a “simple” formula competitors can copy.
- Battery life experiences conflict: some report worse iPhone endurance than Pixels despite Apple’s efficiency claims.
Tooling, narration, and UX
- Related profiling and shader‑optimization videos are recommended.
- Mixed reactions to narration: some find it amateurish or distracting, others fine or even engaging; speed (1.5×) and non‑native accent are mentioned.
- Subtitles contain small transcription errors; surprising to some given Apple’s control.
External display issues
- A recurring complaint: M1/M2 external‑monitor flicker, often tied to color mode (RGB vs other).
- Workarounds exist but are fragile; some blame Apple’s drivers, others note it’s a separate display block from the GPU. Whether M3 fixes this is unclear.